Hugging Face and arXiv Integration for Machine Learning Demos

Hugging Face has collaborated with arXiv to integrate Hugging Face Spaces into arXivLabs. This integration allows users to access interactive, open-source machine learning demos created by authors or the community directly from a paper's arXiv abstract page via a dedicated "Demos" tab.

Direct Access to Interactive Demos on arXiv

The integration between Hugging Face Spaces and arXivLabs enables users to find and launch interactive demos for specific research papers on their respective arXiv abstract pages. By navigating to the Demos tab, users can immediately try out the models described in the paper without needing to write or execute code.

For instance, the BERT language model's arXiv page now features a Demos tab containing over 200 community-built demos. These range from simple showcases of the BERT model itself to complex applications that integrate BERT into larger pipelines.

The Role of Hugging Face Spaces

Hugging Face Spaces allows users to build and share interactive applications that enable anyone with a web browser to experience machine learning models. Since its launch in October 2021, the platform has hosted over 12,000 open-source machine learning demos.

These demos are typically constructed using open-source Python libraries such as Gradio and Streamlit, leveraging models and datasets hosted on the Hugging Face Hub.

Impact on Research Accessibility and Reproducibility

Interactive demos increase the accessibility of computational models across various fields, including biology, chemistry, astronomy, and economics. This integration provides several key benefits to the research community:

  • Increased Visibility: Demos amplify the visibility of researchers' work and make it more discoverable.
  • Improved Understanding: Interactive tools help a wider audience understand how models work and identify or debug biases and other issues.
  • Increased Reproducibility: By allowing users to explore research results without writing code, demos lower the barrier to entry for verifying and reproducing the paper's results.

Sources